When a user wishes to find information on the World Wide Web, he or she may enter a query in a search engine. In response to each query, the search engine may return two types of results: organic (also known as natural) search results and paid search results. Organic search results are those listings that the search engine shows without direct compensation from a third party. Paid search results are advertisements that are only shown so long as the advertiser pays the search engine. Paid search results are often labeled as “sponsored ads,” “sponsored links” or “sponsored results.” The paid search results often appear adjacent to or above the organic search results, but may appear anywhere on the search engine results page (SERP). By way of example, FIG. 1 and FIG. 2 show two SERPs, including organic and paid results
Search engines provide each advertiser with a great deal of control over where the advertisers' ads appear and where they do not appear. Returning to FIG. 1, the company “United Auto Body and Paint” may bid on search terms such as “united,” “automobile,” “collision repair” and/or “united auto body and paint”. Many search engines allow the advertiser to pay a greater amount per impression or per click on an advertisement in order for the advertisement to appear in a better position on the page (e.g., a higher position in the list of paid search results). The advertiser can specify that each ad be shown to users located anywhere in the world or restricted to specific geographic regions, such as only users in San Diego, Calif.
Due to finite budgets, no advertiser can afford to have their listing appear on every SERP. Instead, advertisers typically examine the actions taken by users and only show their ads in scenarios where the return on advertising spend (ROAS) is sufficient. The ROAS is defined as the value of the actions (such as purchasing a product, viewing a webpage, or downloading a white paper) taken by users as a result of a set of advertisements divided by the cost of those advertisements. Similar metrics include return on investment (ROI), cost per acquisition (CPA), cost per success event, cost per value point, and expense to revenue ratio (E/R). Data about the number of impressions served of each search advertisement and the cost of these advertisements is obtained from the search engines such as Google, Yahoo, and MSN. Data about the actions taken by users (conversion data) is obtained from “Web analytics” systems that track usage of the advertiser website. Leading Web analytics products include Coremetrics, Google Analytics, Omniture SiteCatalyst, Unica Netlnsight, and WebTrends Marketing Lab.
Currently, if a user clicks on an advertisement and then takes actions on the advertiser's webpage, most advertisers attribute the value of the actions taken to that advertisement. Returning to FIG. 2, if a user clicks on the “Economist.com/subscribe” advertisement and purchases a subscription on the target website, The Economist probably considers this revenue to result from the advertisement. Supposing that The Economist pays $1 per click on the advertisement, 10% of users who click on the advertisement actually purchase a subscription, and The Economist's net income per additional subscription is $40, The Economist would calculate the ROAS as 10%*$40/$1=4.0. However, a basic premise of the above calculation is flawed because the revenue does not necessarily result from the advertisement.
Therefore, it would be advantageous to understand the true value of an advertisement and to communicate that value to an advertiser.
Moreover, assessment of the impact of various advertising campaigns and programs such as television, magazine, online display ads, and search engine ads is commonly carried out via user surveys and other techniques such as marketing mix modeling. These techniques may be able to assess the effects of major campaigns on a company's key performance metrics such as revenue or website visits, but they cannot handle more granular marketing actions, such as the impact of showing an ad in response to a particular search engine query. Marketing mix modeling typically utilizes two to three years of historical data in the statistical analysis and is generally not used to assess the effect of a routine marketing change implemented for a period of less than one day to two weeks. Finally, surveys and marketing mix modeling both rely heavily on human expertise and custom analysis and do not appear to be well suited to automation.
Therefore, it would be advantageous to automate the assessment of the impacts of various advertising campaigns and programs. Moreover, it would be advantageous to assess impacts of various advertising campaigns using smaller data sets and/or real-time or recent data.